the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 23 Oct 2023
Bayesian Inference-Based Estimation of Hourly Primary and Secondary Organic Carbon at Suburban Hong Kong: Multi-temporal Scale Variations and Evolution Characteristics during PM2.5 episodes
Abstract. Observation-based data of primary and secondary organic carbon in ambient particulate matter (PM) are essential for model evaluation, climate and air quality research, health effects assessment, and mitigation policy development. Since there are no direct measurement tools available to quantify primary organic (POC) and secondary organic carbon (SOC) as separate quantities, their estimation relies on inference approaches using relevant measurable PM constituents. In this study, we measured hourly carbonaceous components and major ions in PM2.5 for a year and a half in suburban Hong Kong from July 2020 to December 2021. We differentiated POC and SOC using a novel Bayesian inference approach, with sulfate identified as the most suitable SOC tracer. The hourly POC and SOC data allowed us to examine temporal characteristics varying from diurnal and weekly patterns to seasonal variations, as well as their evolution characteristics during individual PM2.5 episodes. A total of 65 city-wide PM2.5 episodes were identified throughout the entire studied period, with SOC contributions during individual episodes varying from 10 % to 66 %. In summertime typhoon episodes, elevated SOC levels were observed during daytime hours, and high temperature and NOx levels were identified as significant factors contributing to episodic SOC formation. Winter haze episodes exhibited high SOC levels, likely due to persistent influences from regional transport originating from the northern region to the sampling site. Enhanced SOC formation was observed with the increase in nocturnal NO3 radical (represented by [NO2][O3]) under conditions of high water content and strong acidity. This suggests that aqueous-phase reactions involving NO3 radical were likely a notable contributor to SOC formation during winter haze episodes. The methodology employed in this study for estimating POC and SOC provides practical guidance for other locations with similar monitoring capabilities in place. The availability of hourly POC and SOC data is invaluable for evaluating and improving atmospheric models, as well as understanding the evolution processes of PM pollution episodes. This, in turn, leads to more accurate model predictions and a better understanding of the contributing sources and processes.
- Preprint
(3608 KB) - Metadata XML
-
Supplement
(4442 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-2286', Anonymous Referee #1, 17 Nov 2023
Wang et al. utilize a recently developed new technique in which Bayesian Inference is used to apportion secondary organic carbon (SOC) and primary organic carbon (POC) from a semi continuous OC/EC analyzer and inorganic aerosol for ~1.5 years worth of observations collected at a suburban-rural site near Hong Kong. The potential value in this approach and observations is that the observations are in-situ (no filter collection and potential biases associated) and provide apportionment of organic carbon (OC) to primary vs secondary, which is important for understanding sources and chemistry and providing recommendations for regulations. The authors argue the Bayesian Inference, which is different than the typical approaches used for OC/EC to apportion SOC and POC, is more robust as there is less influence from assumptions.
With these observations, Wang et al. finds that during the 1.5 year study period, most of the OC was from POC. Further, they observe a time of day and day of week influence on POC and SOC, highlighting the sources and/or chemistry that leads to these two components. The authors then investigated the role meteorology on SOC and POC and observed some influence with temperature and RH, depending on time period and/or amount of total aerosol. Further, the authors find different correlations of SOC and POC to other pollutants, potentially highlighting the chemistry or sources of SOC and POC. Finally, the authors look at different influences on SOC and POC--pollution periods, typhoon, and winter.The paper is of potential interest to the ACP community. However, as written, it sometimes appears either more of a methods paper or a measurements report paper. This is partially that the authors just look at correlations of data without providing further context in what that correlation (or lack thereof) means for POC and SOC observations. Further, there is some questions about the results that need to be further explored/discussed to understand the robustness of the technique. Some of these comments are discussed in more detail below and should be addressed prior to publication.
Major
1) The major concern is the premise that sulfate was identified to be most suitable for an SOC tracer and the SOC apportionment for the following reasons that need to be explored and/or addressed:
a) The gas-phase production of sulfuric acid is slow. Even at extremely high OH concentration (~1e7 molec. cm^3 OH), the lifetime of SO2 is ~30 hours, which is slower than the typical production rates observed for fresh SOC production. Further, if there are minimal sources of SO2 in the near-field (e.g., if there is no coal-fire plants, high sulfur diesel, etc.), the combination of this and relatively long provides concern in using sulfate to partition the OC to SOC. Specifically, could it be underestimating the SOC due to the differences in time scale. Could this potentially explain the under performance of the Bayesian Inference method vs PMF during spring (and how does it look for summer and autumn, if possible?).
b) Another important source of sulfate is through aqueous chemistry (typically the most important source of sulfate in continental and polluted regions). As the time and processing of sulfate is different than fast gas-phase production SOC, how is this influencing the results?
c) Due to a) and b), sulfate is more typically considered a "regional" aerosol pollutant and less of a local pollutant. Thus, how much is the partitioning is just assign highly-aged, regional SOC instead of the fresher, rapidly produced SOC?
d) Looking at the paper that describes the method (Liao et al., ES&T, 2023) and the comparisons of the Bayesian Inference partitioning vs the other methods and the results in the SI, the agreement between the Bayesian Inference and PMF is not great, especially in spring. What is leading to this larger disagreement (e.g., have the authors looked at the curvature and where the data is further away from the 1:1 line)? What is the R^2 for these comparisons and slope? Could it be that there are time periods where other a priori aerosol are needed to help infer SOC instead of always using sulfate? Further, the results in the SI do not quite match the good agreement (though no statistical analysis is provided to support this) in line 223, page 7.
e) Part of this relates to the POC > SOC for most of the observations. It is currently unclear if this is due to being near a construction site or mis-apportioning the rapid, fresher SOC to POC. As the authors describe the location being suburban-to-rural, the influence of POC should be lower as there should be lower sources of POC, except for the construction site located right there. If the apportionment can be shown to be robust (e.g., not missing rapidly produced, fresh SOC) then more discussion on the potential influence of the very localized emissions of POC on the measurements and conclusions about SOC vs POC may need to be discussed (e.g., where were the measurements in relation to the construction and did the influence change with wind direction, etc.).2) Page 6, Line 182--It is not clear where the 24x4 groups originate from and the meaning/usage of this. It appears that the 24 comes from the 24 hours in the day, but where is the 4 coming from?
3) It is unclear in Figure 2 what "Corrected" PM2.5 means. What was corrected and why was it corrected? This needs to be described in detail and provide evidence in why the correction was needed to be applied. Further, with figure 2, the color selection for (b) was confusing between the box and whisker plots and the wind direction. As there is overlap, it appears that it someone needs to be interpreted that wind direction corresponds to the different colors.
4) Page 10, line 293: It is surprising that SOC would be more influenced by biogenics. How far are the measurements from Hong Kong and how often is urban outflow from Hong Kong influencing the region?
5) Section 3.3 vs Section 3.4.3 and 3.4.4 are confusing as they appear to highlight different conclusions. More time needs to be used in trying to bring these sections into a more comprehensive story that provides supporting evidence is needed. Section 3.3 is an example where more in-depth discussions are needed (e.g., why there is less correlation of SOC with oxidants). Further, the discussion of the relationship between NOx and SOC needs further investigation, as either the NOx is directly from the construction site, NOx is an indicator of OH, or NOx could actually inhibit the production of SOC as NOx generally leads to more volatile gas-phase products and not less-volatile gas-phase products typically associated with SOC. An example of further discussion/investigation is line 454 - 456 in saying that more NOx = more SOC and that NOx play a role in SOC formation. What role and why?
6) Figure 6--the mass increment ratio needs to be defined in the caption so that the reader does not need to try to find it in the main text.
7) Section 3.4.3--it was surprising at first to see typhoon related episodes led to higher PM and more stagnant conditions when it was typhoon season and the typhoons did not pass through the site. This should be discussed or rephrased as the title of the subsection and the low winds and the PM2.5 seem contradictory.
8) Overall statistics. There are many instances where it was stated there was a good correlation between different variables when the Pearson's coefficient was 0.5 or lower, which means that only ~25% of the variation could be capture by the correlation. Thus, much of the discussion about good correlations needs to be softened or rephrased. Further, it is discussed that there are differences between observations, but looking at the box and whisker plots, the data looks nearly identical. Statistical analysis of the different binned data and including an asterisk for data that is significantly different from other bins would be beneficial. As stated above, Fig. S2 needs more discussion and consideration for statistics. Finally, it is not clear why the units were selected as they were for Fig. S8 (why ln(SOC), ln(RH), and 1000/T))?
9) Section 3.4.4--It is unclear if the necessary compounds were measured to discuss acid-catalyzed reactions (e.g., were gas-phase HNO3 and/or NH3 measured, were the cations listed for ISORROPIA measured)? If none of these were measured, acidity of the aerosol should not be considered as the thermodynamic model is under-constrained. If they were measured, comparisons of the measurements, including the fractional contribution of either gas-phase to its total (e.g., gas-phase HNO3 / (aerosol-phase NO3 + gas-phase NO3)) from the model and observations needs to be provided to provide confidence in the acidity calculations. Further, it is surprising to discuss that there is aqueous-phase nitrate radical chemistry, as the radical has very low Henry's Law constant meaning it does not readily go into the aerosol liquid water. Finally, the last sentence that states that acid-catalyzed reactions with NO3 radical is very surprising as this seems thermodynamically unfavorable. Are there laboratory studies that have observed this? If so, please provide appropriate references. Another concern is that there is correlation of SOC with Ox at nighttime, as O3 should be low at nighttime due to no production. Why is there correlation?
Citation: https://doi.org/10.5194/egusphere-2023-2286-RC1 - AC1: 'Reply on RC1', Jian Zhen Yu, 03 Mar 2024
-
RC2: 'Comment on egusphere-2023-2286', Anonymous Referee #2, 21 Jan 2024
Review of “Bayesian Inference-Based Estimation of Hourly Primary and Secondary Organic Carbon at Suburban Hong Kong: Multi-temporal Scale Variations and Evolution Characteristics during PM2.5 episodes” by Wang et al.
General comments
The manuscript focuses on understanding primary and secondary organic carbon (POC and SOC) in PM2.5 and their driving factors. The study was conducted in suburban Hong Kong from July 2020 to December 2021. It employs a novel Bayesian inference approach to differentiate between POC and SOC, using sulfate as a tracer for SOC. The study explores the temporal characteristics of POC and SOC, including diurnal, weekly, and seasonal variations, and their evolution during PM2.5 episodes. The methodology developed offers practical guidance for similar studies elsewhere, providing valuable insights for atmospheric models and understanding PM pollution processes. The study’s results indicate distinct SOC formations under different seasonal and pollution conditions, influenced by factors like temperature, relative humidity, and atmospheric oxidants. This research contributes to refining atmospheric models and developing strategies for air quality improvement and climate change mitigation. The results are presented effectively, with appropriate statistical analysis and visual aids. This paper is within the scope of ACP and might be of great interest to the broad atmospheric science community.
However, there are areas for improvement in terms of clarity and depth in the discussion of certain results, particularly the implications of the findings for broader atmospheric science and policy-making. Currently, it tends to read more as a data-centric measurement report. Further elaboration on the Bayesian inference method used is recommended for accessibility to readers less familiar with this approach. I agree with the concerns raised by Anonymous Referee #1, particularly regarding the use of sulfate as the best tracer for SOC for all seasons and pollution episodes. Additionally, I have a few specific questions and comments that should also be addressed before the manuscript can be considered for publication.
Specific comments:
1. Figure 1b: Could you explain the noticeable decrease in K1 observed between 6 and 7 am during winter, spring, and fall?
2. Figure 2a: Please clarify the term “corrected PM2.5.” What does this correction entail?
3. Figure S3: The error bar (uncertainty) in Figure S3a should be defined for clarity. Additionally, in Figure S3b, the legend should be corrected to read “SOC_u/SOC_c”.
4. Line 290: Based on Figure 3, it appears that SOC values are slightly higher on weekends, particularly in winter and spring. Could you provide any insights into this observation?
5. Line 298: The reference to daily ozone patterns appears to be incorrectly cited as Figure S4b; it should be Figure 3b.
6. Line 372: The term “ensuring analysis” seems unclear to me. Could you provide a more detailed explanation or rephrase it for clarity?
7. Figure 6: For the non-episode data, is the representation limited to the hours at the start of each season, or does it encompass all non-episode hours throughout the respective season?
Citation: https://doi.org/10.5194/egusphere-2023-2286-RC2 - AC2: 'Reply on RC2', Jian Zhen Yu, 03 Mar 2024
Status: closed
-
RC1: 'Comment on egusphere-2023-2286', Anonymous Referee #1, 17 Nov 2023
Wang et al. utilize a recently developed new technique in which Bayesian Inference is used to apportion secondary organic carbon (SOC) and primary organic carbon (POC) from a semi continuous OC/EC analyzer and inorganic aerosol for ~1.5 years worth of observations collected at a suburban-rural site near Hong Kong. The potential value in this approach and observations is that the observations are in-situ (no filter collection and potential biases associated) and provide apportionment of organic carbon (OC) to primary vs secondary, which is important for understanding sources and chemistry and providing recommendations for regulations. The authors argue the Bayesian Inference, which is different than the typical approaches used for OC/EC to apportion SOC and POC, is more robust as there is less influence from assumptions.
With these observations, Wang et al. finds that during the 1.5 year study period, most of the OC was from POC. Further, they observe a time of day and day of week influence on POC and SOC, highlighting the sources and/or chemistry that leads to these two components. The authors then investigated the role meteorology on SOC and POC and observed some influence with temperature and RH, depending on time period and/or amount of total aerosol. Further, the authors find different correlations of SOC and POC to other pollutants, potentially highlighting the chemistry or sources of SOC and POC. Finally, the authors look at different influences on SOC and POC--pollution periods, typhoon, and winter.The paper is of potential interest to the ACP community. However, as written, it sometimes appears either more of a methods paper or a measurements report paper. This is partially that the authors just look at correlations of data without providing further context in what that correlation (or lack thereof) means for POC and SOC observations. Further, there is some questions about the results that need to be further explored/discussed to understand the robustness of the technique. Some of these comments are discussed in more detail below and should be addressed prior to publication.
Major
1) The major concern is the premise that sulfate was identified to be most suitable for an SOC tracer and the SOC apportionment for the following reasons that need to be explored and/or addressed:
a) The gas-phase production of sulfuric acid is slow. Even at extremely high OH concentration (~1e7 molec. cm^3 OH), the lifetime of SO2 is ~30 hours, which is slower than the typical production rates observed for fresh SOC production. Further, if there are minimal sources of SO2 in the near-field (e.g., if there is no coal-fire plants, high sulfur diesel, etc.), the combination of this and relatively long provides concern in using sulfate to partition the OC to SOC. Specifically, could it be underestimating the SOC due to the differences in time scale. Could this potentially explain the under performance of the Bayesian Inference method vs PMF during spring (and how does it look for summer and autumn, if possible?).
b) Another important source of sulfate is through aqueous chemistry (typically the most important source of sulfate in continental and polluted regions). As the time and processing of sulfate is different than fast gas-phase production SOC, how is this influencing the results?
c) Due to a) and b), sulfate is more typically considered a "regional" aerosol pollutant and less of a local pollutant. Thus, how much is the partitioning is just assign highly-aged, regional SOC instead of the fresher, rapidly produced SOC?
d) Looking at the paper that describes the method (Liao et al., ES&T, 2023) and the comparisons of the Bayesian Inference partitioning vs the other methods and the results in the SI, the agreement between the Bayesian Inference and PMF is not great, especially in spring. What is leading to this larger disagreement (e.g., have the authors looked at the curvature and where the data is further away from the 1:1 line)? What is the R^2 for these comparisons and slope? Could it be that there are time periods where other a priori aerosol are needed to help infer SOC instead of always using sulfate? Further, the results in the SI do not quite match the good agreement (though no statistical analysis is provided to support this) in line 223, page 7.
e) Part of this relates to the POC > SOC for most of the observations. It is currently unclear if this is due to being near a construction site or mis-apportioning the rapid, fresher SOC to POC. As the authors describe the location being suburban-to-rural, the influence of POC should be lower as there should be lower sources of POC, except for the construction site located right there. If the apportionment can be shown to be robust (e.g., not missing rapidly produced, fresh SOC) then more discussion on the potential influence of the very localized emissions of POC on the measurements and conclusions about SOC vs POC may need to be discussed (e.g., where were the measurements in relation to the construction and did the influence change with wind direction, etc.).2) Page 6, Line 182--It is not clear where the 24x4 groups originate from and the meaning/usage of this. It appears that the 24 comes from the 24 hours in the day, but where is the 4 coming from?
3) It is unclear in Figure 2 what "Corrected" PM2.5 means. What was corrected and why was it corrected? This needs to be described in detail and provide evidence in why the correction was needed to be applied. Further, with figure 2, the color selection for (b) was confusing between the box and whisker plots and the wind direction. As there is overlap, it appears that it someone needs to be interpreted that wind direction corresponds to the different colors.
4) Page 10, line 293: It is surprising that SOC would be more influenced by biogenics. How far are the measurements from Hong Kong and how often is urban outflow from Hong Kong influencing the region?
5) Section 3.3 vs Section 3.4.3 and 3.4.4 are confusing as they appear to highlight different conclusions. More time needs to be used in trying to bring these sections into a more comprehensive story that provides supporting evidence is needed. Section 3.3 is an example where more in-depth discussions are needed (e.g., why there is less correlation of SOC with oxidants). Further, the discussion of the relationship between NOx and SOC needs further investigation, as either the NOx is directly from the construction site, NOx is an indicator of OH, or NOx could actually inhibit the production of SOC as NOx generally leads to more volatile gas-phase products and not less-volatile gas-phase products typically associated with SOC. An example of further discussion/investigation is line 454 - 456 in saying that more NOx = more SOC and that NOx play a role in SOC formation. What role and why?
6) Figure 6--the mass increment ratio needs to be defined in the caption so that the reader does not need to try to find it in the main text.
7) Section 3.4.3--it was surprising at first to see typhoon related episodes led to higher PM and more stagnant conditions when it was typhoon season and the typhoons did not pass through the site. This should be discussed or rephrased as the title of the subsection and the low winds and the PM2.5 seem contradictory.
8) Overall statistics. There are many instances where it was stated there was a good correlation between different variables when the Pearson's coefficient was 0.5 or lower, which means that only ~25% of the variation could be capture by the correlation. Thus, much of the discussion about good correlations needs to be softened or rephrased. Further, it is discussed that there are differences between observations, but looking at the box and whisker plots, the data looks nearly identical. Statistical analysis of the different binned data and including an asterisk for data that is significantly different from other bins would be beneficial. As stated above, Fig. S2 needs more discussion and consideration for statistics. Finally, it is not clear why the units were selected as they were for Fig. S8 (why ln(SOC), ln(RH), and 1000/T))?
9) Section 3.4.4--It is unclear if the necessary compounds were measured to discuss acid-catalyzed reactions (e.g., were gas-phase HNO3 and/or NH3 measured, were the cations listed for ISORROPIA measured)? If none of these were measured, acidity of the aerosol should not be considered as the thermodynamic model is under-constrained. If they were measured, comparisons of the measurements, including the fractional contribution of either gas-phase to its total (e.g., gas-phase HNO3 / (aerosol-phase NO3 + gas-phase NO3)) from the model and observations needs to be provided to provide confidence in the acidity calculations. Further, it is surprising to discuss that there is aqueous-phase nitrate radical chemistry, as the radical has very low Henry's Law constant meaning it does not readily go into the aerosol liquid water. Finally, the last sentence that states that acid-catalyzed reactions with NO3 radical is very surprising as this seems thermodynamically unfavorable. Are there laboratory studies that have observed this? If so, please provide appropriate references. Another concern is that there is correlation of SOC with Ox at nighttime, as O3 should be low at nighttime due to no production. Why is there correlation?
Citation: https://doi.org/10.5194/egusphere-2023-2286-RC1 - AC1: 'Reply on RC1', Jian Zhen Yu, 03 Mar 2024
-
RC2: 'Comment on egusphere-2023-2286', Anonymous Referee #2, 21 Jan 2024
Review of “Bayesian Inference-Based Estimation of Hourly Primary and Secondary Organic Carbon at Suburban Hong Kong: Multi-temporal Scale Variations and Evolution Characteristics during PM2.5 episodes” by Wang et al.
General comments
The manuscript focuses on understanding primary and secondary organic carbon (POC and SOC) in PM2.5 and their driving factors. The study was conducted in suburban Hong Kong from July 2020 to December 2021. It employs a novel Bayesian inference approach to differentiate between POC and SOC, using sulfate as a tracer for SOC. The study explores the temporal characteristics of POC and SOC, including diurnal, weekly, and seasonal variations, and their evolution during PM2.5 episodes. The methodology developed offers practical guidance for similar studies elsewhere, providing valuable insights for atmospheric models and understanding PM pollution processes. The study’s results indicate distinct SOC formations under different seasonal and pollution conditions, influenced by factors like temperature, relative humidity, and atmospheric oxidants. This research contributes to refining atmospheric models and developing strategies for air quality improvement and climate change mitigation. The results are presented effectively, with appropriate statistical analysis and visual aids. This paper is within the scope of ACP and might be of great interest to the broad atmospheric science community.
However, there are areas for improvement in terms of clarity and depth in the discussion of certain results, particularly the implications of the findings for broader atmospheric science and policy-making. Currently, it tends to read more as a data-centric measurement report. Further elaboration on the Bayesian inference method used is recommended for accessibility to readers less familiar with this approach. I agree with the concerns raised by Anonymous Referee #1, particularly regarding the use of sulfate as the best tracer for SOC for all seasons and pollution episodes. Additionally, I have a few specific questions and comments that should also be addressed before the manuscript can be considered for publication.
Specific comments:
1. Figure 1b: Could you explain the noticeable decrease in K1 observed between 6 and 7 am during winter, spring, and fall?
2. Figure 2a: Please clarify the term “corrected PM2.5.” What does this correction entail?
3. Figure S3: The error bar (uncertainty) in Figure S3a should be defined for clarity. Additionally, in Figure S3b, the legend should be corrected to read “SOC_u/SOC_c”.
4. Line 290: Based on Figure 3, it appears that SOC values are slightly higher on weekends, particularly in winter and spring. Could you provide any insights into this observation?
5. Line 298: The reference to daily ozone patterns appears to be incorrectly cited as Figure S4b; it should be Figure 3b.
6. Line 372: The term “ensuring analysis” seems unclear to me. Could you provide a more detailed explanation or rephrase it for clarity?
7. Figure 6: For the non-episode data, is the representation limited to the hours at the start of each season, or does it encompass all non-episode hours throughout the respective season?
Citation: https://doi.org/10.5194/egusphere-2023-2286-RC2 - AC2: 'Reply on RC2', Jian Zhen Yu, 03 Mar 2024
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 312 | 105 | 22 | 439 | 38 | 11 | 12 |
- HTML: 312
- PDF: 105
- XML: 22
- Total: 439
- Supplement: 38
- BibTeX: 11
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1